arxiv: 2602.22507 · v2 · submitted 2026-02-26 · 💻 cs.LG · cs.CV

Recognition: no theorem link

Space Syntax-guided Post-training for Residential Floor Plan Generation

Zhuoyang Jiang , Dongqing Zhang

Authors on Pith no claims yet

Pith reviewed 2026-05-15 19:36 UTC · model grok-4.3

classification 💻 cs.LG cs.CV

keywords space syntaxfloor plan generationpost-trainingreinforcement learningconfigurational qualityresidential layouts

0 comments

The pith

Space syntax integration scores can be used as post-training feedback to improve configurational quality in generated residential floor plans.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Residential floor plan generators often produce layouts that are geometrically plausible yet fail to place shared spaces centrally and private ones apart. The paper turns space syntax analysis into an optimization signal by building an oracle that scores generated rectangle layouts on public-space dominance and functional hierarchy. This oracle first calibrates against real residential data, then drives two post-training routes: iterative generate-filter-retrain and reinforcement learning via PPO. Both routes raise alignment with observed real-world patterns, and the PPO route delivers larger gains, lower variance, and greater efficiency. The result is a practical way to inject output-side architectural evaluation into already-trained generators.

Core claim

The authors establish that the Space Syntax Integration Oracle converts generated rectangle layouts into graphs and supplies integration scores that serve as actionable targets for post-training, producing measurable gains in public-space dominance and functional-hierarchy alignment over unpost-trained baselines, with the PPO strategy outperforming iterative retraining.

What carries the argument

The Space Syntax Integration Oracle (SSIO), which converts generated layouts into rectangle-space graphs and computes integration values to quantify public-space dominance and functional hierarchy.

If this is right

Post-trained generators produce layouts whose configurational statistics align more closely with empirical references from real residential data.
The PPO-based post-training route achieves larger gains, lower variance, and higher efficiency than iterative generate-filter-retrain.
Output-side configurational evaluation can function as effective feedback for existing floor plan generation backbones.
Architectural theory can be injected into generative models through post-training rather than input-side conditioning alone.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same oracle-driven feedback loop could be tested on other building types whose spatial logic is also governed by accessibility hierarchies.
Combining the post-training signal with existing room-relation graph inputs might compound the observed improvements.
Whether higher oracle scores translate into measurable differences in occupant navigation or satisfaction remains an open empirical question.

Load-bearing premise

Space syntax integration scores computed on simplified rectangle layouts reliably indicate desirable configurational qualities in actual residential buildings.

What would settle it

A controlled comparison of architect or resident ratings for layouts that score high versus low on the Space Syntax Integration Oracle.

Figures

Figures reproduced from arXiv: 2602.22507 by Dongqing Zhang, Zhuoyang Jiang.

**Figure 1.** Figure 1: Overview Framework to bridge the gap between generative probability and spatial logic. The overview framework is in [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗

**Figure 2.** Figure 2: Overall relative integration pattern across room types (a) Functional-domain–level harmonized comparison (b) Cross-dataset validation of public-space dominance [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

**Figure 3.** Figure 3: Dataset-specific room-type coverage and labeling differences plans, and this empirically grounded hierarchy motivates the use of living/public-space integration as a robust design prior for guiding early-stage floor plan generation. 3.2.2. Quality Evaluation of Baseline Models and Initial Model Selection To evaluate the spatial organization of AI-generated residential floor plans against empirical architec… view at source ↗

**Figure 4.** Figure 4: compares the coverage-weighted relative integration profiles of the screened RPLAN dataset and 10,000 residential floor plans generated by HouseDiffusion (HD10000), one of the most stable and advanced AI-based residential floor plan generation models to date. For both datasets, integration values are normalized within each plan, and the baseline value of 1 represents the unweighted per-plan average. It s… view at source ↗

**Figure 5.** Figure 5: SSPT-Bench (Eval-8) median trends over post-training iterations. SSPT-PPO consistently improves public-space dominance and living-room advantage, while producing a notably narrower (more stable) distribution for living-room relative integration. (a) public_score (IQR band). (b) living_adv (IQR band). (c) living_room (IQR band) [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗

**Figure 6.** Figure 6: Robust dispersion comparison under SSPT-Bench (Eval-8). IQR bands show that SSPT-PPO yields not only higher medians on public_score and living_adv, but also consistently tighter distributions, indicating improved controllability and reduced sensitivity to sampling noise. : Preprint submitted to Elsevier Page 16 of 20 [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗

**Figure 7.** Figure 7: Room-type relative integration profile evolution under SSPT-Bench. SSPT-PPO yields a closer overall match to the empirical hierarchy in RPLAN8 , particularly by reducing over-integration of non-public categories (e.g., Entrance/Bedroom/Bathroom), which increases living-room dominance and advantage without requiring a large increase in the absolute living-room level. be applied to other generators (e.g., t… view at source ↗

read the original abstract

Residential floor plan generation requires not only geometric fidelity but also spatial configurational logic: shared living spaces should be integrative, while private spaces should remain segregated. Existing generators increasingly use room-relation graphs as input-side conditions, but generated layouts are rarely evaluated on the output side for configurational quality, and such evaluation is rarely fed back into model optimization. We propose Space Syntax-guided Post-training (SSPT), a framework that turns space-syntax integration from a post-hoc analysis tool into a computable feedback signal for already-trained floor plan generators. SSPT introduces the Space Syntax Integration Oracle (SSIO), which converts generated layouts into rectangle-space graphs and measures public-space dominance and functional hierarchy. SSIO is first applied to real residential data to establish empirical configurational references, then connected to two SSPT strategies: SSPT-Iter, a basic generate-filter-retrain route, and SSPT-PPO, the first RL-based post-training route for floor plan generation. We also introduce SSPT-Bench, a new evaluation system for measuring the output-side spatial configurational quality of post-trained generators under an out-of-distribution setting. Experiments show that both strategies improve public-space dominance and functional-hierarchy alignment over the unpost-trained baseline. SSPT-PPO achieves stronger gains, lower variance, and higher efficiency than iterative retraining. These results show that output-side configurational evaluation can serve as actionable post-training feedback, offering a practical path for injecting architectural theory into existing floor plan generation backbones.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

SSPT turns space syntax into a post-training feedback loop for floor plans, with PPO outperforming iteration, though experimental details are sparse.

read the letter

The main point to take away is that the authors have turned space syntax integration into a usable feedback signal for post-training floor plan generators. They introduce the SSIO to compute public-space dominance and functional hierarchy from generated rectangle layouts, then apply it in two ways: a simple iterative generate-filter-retrain loop and a PPO-based RL approach. The PPO version reportedly gives stronger, more stable improvements. This is new in the way it closes the loop from output evaluation back to optimization using architectural theory. Prior work conditions on graphs at input but rarely optimizes the output configurational properties this way. Setting up SSPT-Bench for OOD testing is also a solid addition. The results show gains on the target metrics, and the efficiency of PPO over iteration is a practical plus. The soft spots are mostly around the strength of the evidence. The abstract and available details don't include enough on baseline models, dataset sizes, or how the out-of-distribution splits were made, so it's difficult to judge how general the improvements are. There's also the open question of whether optimizing for SSIO scores on simplified rectangle graphs actually leads to better real-world floor plans, or if it just tunes to the metric without independent confirmation like user studies. Overall, this is for people in computational design and generative AI for buildings. If you're looking at ways to make generated plans more architecturally sensible, the post-training framework is worth examining. It should go to peer review because the core mechanism is clearly described and builds on solid prior concepts, though the experiments will likely need expansion.

Referee Report

3 major / 2 minor

Summary. The paper proposes Space Syntax-guided Post-training (SSPT) to improve configurational quality in residential floor plan generators. It introduces the Space Syntax Integration Oracle (SSIO) that converts generated rectangle layouts into graphs and computes scores for public-space dominance and functional hierarchy. These scores are used as feedback in two post-training strategies (SSPT-Iter and SSPT-PPO) and evaluated on a new SSPT-Bench under out-of-distribution conditions, with experiments claiming that both methods improve alignment over the baseline and that SSPT-PPO yields stronger, lower-variance gains.

Significance. If the central claims hold after addressing validation gaps, the work provides a concrete mechanism for injecting established architectural theory (space syntax) into existing ML generators via post-training rather than retraining from scratch. The RL-based SSPT-PPO route and the SSPT-Bench evaluation protocol are potentially useful contributions for the floor-plan generation community. The significance is currently limited by insufficient experimental detail and lack of independent checks on whether SSIO scores reliably proxy desirable real-world configurational properties.

major comments (3)

[Experimental Evaluation] Experimental Evaluation section: The reported gains (stronger performance, lower variance, higher efficiency for SSPT-PPO) are presented without sample sizes, statistical significance tests, exact baseline architectures, or details on how the out-of-distribution splits in SSPT-Bench were constructed. These omissions prevent assessment of whether the improvements are robust or reproducible.
[§3] SSIO definition (§3): No independent validation is provided (e.g., correlation with human expert ratings, alternative metrics such as visibility graphs, or failure-case analysis) showing that SSIO scores computed on rectangle graphs correspond to desirable real-world configurational quality rather than artifacts of the rectangle approximation or the oracle itself.
[SSPT-Bench] SSPT-Bench and reference construction: Because the same SSIO is used both to derive empirical references from real data and as the optimization target/feedback signal, it is unclear whether observed OOD improvements reflect genuine generalization of configurational logic or optimization to metric-specific properties; an ablation or cross-metric check is needed.

minor comments (2)

[§3] The conversion process from generated layouts to rectangle-space graphs is described at a high level; a diagram or pseudocode would improve clarity of the SSIO pipeline.
[Abstract] Notation for the two SSPT strategies (SSPT-Iter vs. SSPT-PPO) should be introduced consistently in the abstract and early sections to avoid reader confusion.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below with specific plans for revision, focusing on improving experimental rigor, validation, and clarity without overstating current results.

read point-by-point responses

Referee: [Experimental Evaluation] Experimental Evaluation section: The reported gains (stronger performance, lower variance, higher efficiency for SSPT-PPO) are presented without sample sizes, statistical significance tests, exact baseline architectures, or details on how the out-of-distribution splits in SSPT-Bench were constructed. These omissions prevent assessment of whether the improvements are robust or reproducible.

Authors: We agree these omissions limit assessment of robustness. In the revised manuscript we will expand the Experimental Evaluation section to report exact sample sizes for all metrics, include statistical significance tests (paired t-tests with p-values and confidence intervals), specify the precise baseline architectures and hyperparameters used, and detail the construction of OOD splits in SSPT-Bench including selection criteria, data partitioning method, and any filtering steps. These additions will be placed in the main text and supplementary material as needed. revision: yes
Referee: [§3] SSIO definition (§3): No independent validation is provided (e.g., correlation with human expert ratings, alternative metrics such as visibility graphs, or failure-case analysis) showing that SSIO scores computed on rectangle graphs correspond to desirable real-world configurational quality rather than artifacts of the rectangle approximation or the oracle itself.

Authors: We acknowledge the need for stronger validation of SSIO. The manuscript grounds SSIO in established space-syntax theory and empirical references derived from real residential data. In revision we will add to §3 a failure-case analysis together with quantitative comparisons against an alternative metric (visibility-graph integration). New human-expert rating collection lies outside the feasible scope of this revision cycle; we will therefore add an explicit limitations paragraph noting this gap and suggesting it for future work. We maintain that the current empirical grounding provides a defensible starting point but accept that deeper independent checks are required. revision: partial
Referee: [SSPT-Bench] SSPT-Bench and reference construction: Because the same SSIO is used both to derive empirical references from real data and as the optimization target/feedback signal, it is unclear whether observed OOD improvements reflect genuine generalization of configurational logic or optimization to metric-specific properties; an ablation or cross-metric check is needed.

Authors: We agree this circularity concern must be addressed. We will add an ablation experiment to the revised SSPT-Bench evaluation that measures post-training gains using an independent configurational metric (visibility-graph analysis) not derived from SSIO. Results will be reported alongside the original SSIO-based scores to demonstrate whether improvements persist under cross-metric evaluation, thereby supporting claims of genuine generalization rather than metric-specific overfitting. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper defines SSIO as a new oracle that converts rectangle layouts to graphs and computes integration values drawn from established space-syntax literature. References are built by applying SSIO to real residential data, then used as targets for two post-training strategies whose gains are measured experimentally on SSPT-Bench. No equation reduces a claimed prediction to a fitted parameter by construction, no load-bearing premise rests on self-citation, and the central empirical claim (improved public-space dominance and hierarchy alignment) is not definitionally equivalent to the input metrics. The derivation therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The central claim rests on the assumption that space-syntax integration values computed on rectangle graphs are valid proxies for human-perceived configurational quality. No free parameters are explicitly named, but empirical references are derived from real residential data. No new physical entities are postulated.

axioms (1)

domain assumption Space-syntax integration metrics computed on rectangle-space graphs accurately reflect desirable public-private hierarchy in residential layouts.
Invoked when SSIO is used to establish empirical references and as optimization target.

invented entities (2)

Space Syntax Integration Oracle (SSIO) no independent evidence
purpose: Converts generated layouts into graphs and computes public-space dominance and functional hierarchy scores.
New component introduced to turn post-hoc analysis into training feedback.
SSPT-Bench no independent evidence
purpose: Evaluation system for output-side configurational quality under out-of-distribution setting.
New benchmark introduced alongside the method.

pith-pipeline@v0.9.0 · 5562 in / 1391 out tokens · 18965 ms · 2026-05-15T19:36:23.171843+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages · 3 internal anchors

[1]

Architectural layout generation using a graph-constrained conditional generative adversarial network (gan)

Aalaei, M., Saadi, M., Rahbar, M., Ekhlassi, A., 2023. Architectural layout generation using a graph-constrained conditional generative adversarial network (gan). Automation in Construction 155, 105053

work page 2023
[2]

The applica- tion of space syntax to enhance sociability in public urban spaces: A systematic review

Askarizad, R., Lamíquiz Daudén, P.J., Garau, C., 2024. The applica- tion of space syntax to enhance sociability in public urban spaces: A systematic review. ISPRS International Journal of Geo-Information

work page 2024
[3]

URL:https://www.mdpi.com/2220-9964/13/7/227, doi:10.3390/ ijgi13070227

work page
[4]

The complexity of testing all properties of planar graphs, and the role of isomorphism

Basu, S., Kumar, A., Seshadhri, C., 2021. The complexity of testing all properties of planar graphs, and the role of isomorphism. ArXiv abs/2108.10547. URL:https://api.semanticscholar.org/CorpusID: 237278741

work page arXiv 2021
[5]

In-between spaces and social interaction: a morphological analysis of izmir using space syntax

Can, I., Heath, T., 2016. In-between spaces and social interaction: a morphological analysis of izmir using space syntax. Journal of Housing and the Built Environment 31, 31–49. URL:https://api. semanticscholar.org/CorpusID:55678336

work page 2016
[6]

Principles for public space design, planning to do better

Carmona, M., 2018. Principles for public space design, planning to do better. URBAN DESIGN International 24, 47–59. URL: https://api.semanticscholar.org/CorpusID:115439896

work page 2018
[7]

Deep reinforcement learning from human preferences

Christiano, P.F., Leike, J., Brown, T., Martic, M., Legg, S., Amodei, D., 2017. Deep reinforcement learning from human preferences. Advances in neural information processing systems 30

work page 2017
[8]

Adopting the topology optimization in the design of high-speed synchronous reluctancemotorsforelectricvehicles.IEEETransactionsonIndustry Applications 56, 5429–5438

Credo, A., Fabri, G., Villani, M., Popescu, M., 2020. Adopting the topology optimization in the design of high-speed synchronous reluctancemotorsforelectricvehicles.IEEETransactionsonIndustry Applications 56, 5429–5438. doi:10.1109/TIA.2020.3007366

work page doi:10.1109/tia.2020.3007366 2020
[9]

Zillowindoordataset:Annotatedfloorplanswith360deg panoramas and 3d room layouts, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

Cruz,S.,Hutchcroft,W.,Li,Y.,Khosravan,N.,Boyadzhiev,I.,Kang, S.B.,2021. Zillowindoordataset:Annotatedfloorplanswith360deg panoramas and 3d room layouts, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 2133– 2143

work page 2021
[10]

arXiv e-prints , arXiv–2206

Deitke, M., VanderBilt, E., Herrasti, A., Weihs, L., Salvador, J., Ehsani, K., Han, W., Kolve, E., Farhadi, A., Kembhavi, A., et al., 2022.Procthor:Large-scaleembodiedaiusingproceduralgeneration. arXiv e-prints , arXiv–2206

work page 2022
[11]

Relation of domestic space preferences with space syntax parameters

Edgü, E., Ünlü, A., 2003. Relation of domestic space preferences with space syntax parameters. URL:https://api.semanticscholar. org/CorpusID:146403316

work page 2003
[12]

Transforminghousingtypologies.space syntax evaluation and shape grammar generation

Eloy,S.,Guerreiro,R.,2016. Transforminghousingtypologies.space syntax evaluation and shape grammar generation. arq.urb , 86–114

work page 2016
[13]

Dpok: Reinforcement learning for fine-tuning text-to-image diffusion models

Fan, Y., Watkins, O., Du, Y., Liu, H., Ryu, M., Boutilier, C., Abbeel, P., Ghavamzadeh, M., Lee, K., Lee, K., 2023. Dpok: Reinforcement learning for fine-tuning text-to-image diffusion models. Advances in Neural Information Processing Systems 36, 79858–79885

work page 2023
[14]

Completely derandomized self- adaptation in evolution strategies

Hansen, N., Ostermeier, A., 2001. Completely derandomized self- adaptation in evolution strategies. Evolutionary computation 9, 159– 195

work page 2001
[15]

Contents

Hanson, J., 1999. Contents. Cambridge University Press. p. v–v

work page 1999
[16]

iplan: Interactive and procedural layout planning, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

He, F., Huang, Y., Wang, H., 2022. iplan: Interactive and procedural layout planning, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 7793–7802. :Preprint submitted to Elsevier Page 18 of 20

work page 2022
[17]

Spatial configuration: Semi- automatic methods for layout generation in practice

Helme, L., Derix, C., Izaki, Å., 2014. Spatial configuration: Semi- automatic methods for layout generation in practice. The Journal of Space Syntax 5, 35–49. URL:https://api.semanticscholar.org/ CorpusID:59127252

work page 2014
[18]

Space is the machine: A configurational theory of architecture

Hillier, B., 1996. Space is the machine: A configurational theory of architecture. URL:https://api.semanticscholar.org/CorpusID: 60919128

work page 1996
[19]

Hu, R., Huang, Z., Tang, Y., Van Kaick, O., Zhang, H., Huang, H.,

work page
[20]

ACM Transactions on Graphics (TOG) 39, 118–1

Graph2plan:Learningfloorplangenerationfromlayoutgraphs. ACM Transactions on Graphics (TOG) 39, 118–1

work page
[21]

Advancing ar- chitecturalfloorplandesignwithgeometry-enhancedgraphdiffusion

Hu, S., Wu, W., Wang, Y., Xu, B., Zheng, L., 2024. Advancing ar- chitecturalfloorplandesignwithgeometry-enhancedgraphdiffusion. arXiv preprint arXiv:2408.16258 8

work page arXiv 2024
[22]

Integration of space syntax into gis for modelling urban spaces

Jiang, B., Claramunt, C., Klarqvist, B., 2000. Integration of space syntax into gis for modelling urban spaces. Interna- tional Journal of Applied Earth Observation and Geoinformation 2, 161–171. URL:https://www.sciencedirect.com/science/article/ pii/S0303243400850102, doi:https://doi.org/10.1016/S0303-2434(00) 85010-2

work page doi:10.1016/s0303-2434(00 2000
[23]

Cubicasa5k:Adatasetandanimprovedmulti-taskmodelforfloorplan image analysis, in: Scandinavian Conference on Image Analysis, Springer

Kalervo, A., Ylioinas, J., Häikiö, M., Karhu, A., Kannala, J., 2019. Cubicasa5k:Adatasetandanimprovedmulti-taskmodelforfloorplan image analysis, in: Scandinavian Conference on Image Analysis, Springer. pp. 28–40

work page 2019
[24]

Kiyota, Y., 2018. Promoting open innovations in real estate tech: Provision of the lifull home’s data set and collaborative studies, in: Proceedings of the 2018 ACM on International Conference on Multimedia Retrieval, pp. 6–6

work page 2018
[25]

End-to-end graph-constrained vectorized floorplan generation with panoptic refinement, in: European Conference on Computer Vision, Springer

Liu,J.,Xue,Y.,Duarte,J.,Shekhawat,K.,Zhou,Z.,Huang,X.,2022. End-to-end graph-constrained vectorized floorplan generation with panoptic refinement, in: European Conference on Computer Vision, Springer. pp. 547–562

work page 2022
[26]

Floorplangan:Vectorresidentialfloorplan adversarial generation

Luo,Z.,Huang,W.,2022. Floorplangan:Vectorresidentialfloorplan adversarial generation. Automation in construction 142, 104470

work page 2022
[27]

Luo, Z.H., Lara, L., Luo, G.Y., Golemo, F., Beckham, C., Pal, C.,

work page
[28]

arXiv preprint arXiv:2407.15723

Dstruct2design:Dataandbenchmarksfordatastructuredriven generative floor plan design. arXiv preprint arXiv:2407.15723

work page arXiv
[29]

The walkable envi- ronment: a systematic review through the lens of space syntax as an integrated approach

Mehrinejad Khotbehsara, E., Yu, R., Somasundaraswaran, K., Askarizad, R., Kolbe-Alexander, T., 2025. The walkable envi- ronment: a systematic review through the lens of space syntax as an integrated approach. Smart and Sustainable Built Environment doi:10.1108/SASBE-02-2024-0049

work page doi:10.1108/sasbe-02-2024-0049 2025
[30]

Conditional Generative Adversarial Nets

Mirza, M., Osindero, S., 2014. Conditional generative adversarial nets. arXiv preprint arXiv:1411.1784

work page internal anchor Pith review Pith/arXiv arXiv 2014
[31]

Illuminating search spaces by mapping elites

Mouret,J.B.,Clune,J.,2015. Illuminatingsearchspacesbymapping elites. arXiv preprint arXiv:1504.04909

work page internal anchor Pith review Pith/arXiv arXiv 2015
[32]

Nauata,N.,Chang,K.H.,Cheng,C.Y.,Mori,G.,Furukawa,Y.,2020. House-gan: Relational generative adversarial networks for graph- constrained house layout generation, in: Computer Vision – ECCV 2020:16thEuropeanConference,Glasgow,UK,August23–28,2020, Proceedings,Part I,Springer-Verlag, Berlin,Heidelberg.p. 162–177. URL:https://doi.org/10.1007/978-3-030-58452-8_1...

work page doi:10.1007/978-3-030-58452-8_10 2020
[33]

Nauata, N., Hosseini, S., Chang, K.H., Chu, H., Cheng, C.Y., Fu- rukawa, Y., 2021. House-gan++: Generative adversarial layout refinement network towards intelligent computational agent for pro- fessional architects, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 13632–13641

work page 2021
[34]

Training language models to follow instructions with human feedback

Ouyang,L.,Wu,J.,Jiang,X.,Almeida,D.,Wainwright,C.,Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al., 2022. Training language models to follow instructions with human feedback. Ad- vances in neural information processing systems 35, 27730–27744

work page 2022
[35]

6690–6700

Para,W.,Guerrero,P.,Kelly,T.,Guibas,L.J.,Wonka,P.,2021.Gener- ative layout modeling using constraint graphs, in: Proceedings of the IEEE/CVFInternationalConferenceonComputerVision(ICCV),pp. 6690–6700

work page 2021
[36]

Directpreferenceoptimization:Yourlanguagemodel issecretlyarewardmodel.Advancesinneuralinformationprocessing systems 36, 53728–53741

Rafailov, R., Sharma, A., Mitchell, E., Manning, C.D., Ermon, S., Finn,C.,2023. Directpreferenceoptimization:Yourlanguagemodel issecretlyarewardmodel.Advancesinneuralinformationprocessing systems 36, 53728–53741

work page 2023
[37]

A room for living: Private and public as- pects in the experience of the living room

Rechavi, T.B., 2009. A room for living: Private and public as- pects in the experience of the living room. Journal of Environ- mental Psychology 29, 133–143. URL:https://www.sciencedirect. com/science/article/pii/S027249440800042X,doi:https://doi.org/10. 1016/j.jenvp.2008.05.001

work page 2009
[38]

Deep generative models in engineering design: A review

Regenwetter, L., Nobari, A.H., Ahmed, F., 2021. Deep generative models in engineering design: A review. CoRR abs/2110.10863. URL:https://arxiv.org/abs/2110.10863,arXiv:2110.10863

work page arXiv 2021
[39]

Z., Lidard, J., Ankile, L

Ren, A.Z., Lidard, J., Ankile, L.L., Simeonov, A., Agrawal, P., Ma- jumdar, A., Burchfiel, B., Dai, H., Simchowitz, M., 2024. Diffusion policy policy optimization. arXiv preprint arXiv:2409.00588

work page arXiv 2024
[40]

Ross, S., Gordon, G., Bagnell, D., 2011. A reduction of imitation learning and structured prediction to no-regret online learning, in: Proceedings of the fourteenth international conference on artificial intelligenceandstatistics,JMLRWorkshopandConferenceProceed- ings. pp. 627–635

work page 2011
[41]

Schulman, J., Wolski, F., Dhariwal, P., Radford, A., Klimov, O.,

work page
[42]

Proximal Policy Optimization Algorithms

Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347

work page internal anchor Pith review Pith/arXiv arXiv
[43]

Shabani, M.A., Hosseini, S., Furukawa, Y., 2023. Housediffusion: Vector floorplan generation via a diffusion model with discrete and continuous denoising, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 5466–5475

work page 2023
[44]

Beyondhuman data:Scalingself-trainingforproblem-solvingwithlanguagemodels

Singh, A., Co-Reyes, J.D., Agarwal, R., Anand, A., Patil, P., Garcia, X.,Liu,P.J.,Harrison,J.,Lee,J.,Xu,K.,etal.,2023. Beyondhuman data:Scalingself-trainingforproblem-solvingwithlanguagemodels. arXiv preprint arXiv:2312.06585

work page arXiv 2023
[45]

Wallplan:synthesizingfloorplansbylearningtogeneratewallgraphs

Sun, J., Wu, W., Liu, L., Min, W., Zhang, G., Zheng, L., 2022. Wallplan:synthesizingfloorplansbylearningtogeneratewallgraphs. ACM Transactions on Graphics (TOG) 41, 1–14

work page 2022
[46]

Graph transformer gans for graph-constrained house generation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

Tang, H., Zhang, Z., Shi, H., Li, B., Shao, L., Sebe, N., Timofte, R., Van Gool, L., 2023. Graph transformer gans for graph-constrained house generation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 2173–2182

work page 2023
[47]

Upadhyay, A., Dubey, A., Arora, V., Kuriakose, S.M., Agarawal, S.,

work page
[48]

Flnet: graph constrained floor layout generation, in: 2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW), IEEE. pp. 1–6

work page 2022
[49]

Buildinginformationmodelling(bim)application for an existing road infrastructure

Vignali,V.,Acerra,E.M.,Lantieri,C.,DiVincenzo,F.,Piacentini,G., Pancaldi,S.,2021. Buildinginformationmodelling(bim)application for an existing road infrastructure. Automation in Construction 128,103752. URL:https://www.sciencedirect.com/science/article/ pii/S092658052100203X, doi:https://doi.org/10.1016/j.autcon.2021. 103752

work page doi:10.1016/j.autcon.2021 2021
[50]

Wang, Y., Kordi, Y., Mishra, S., Liu, A., Smith, N.A., Khashabi, D., Hajishirzi, H., 2023. Self-instruct: Aligning language models with self-generated instructions, in: Proceedings of the 61st annual meeting of the association for computational linguistics (volume 1: long papers), pp. 13484–13508

work page 2023
[51]

Data-driven interior plan generation for residential buildings

Wu, W., Fu, X.M., Tang, R., Wang, Y., Qi, Y.H., Liu, L., 2019. Data-driven interior plan generation for residential buildings. ACM Trans. Graph. 38. URL:https://doi.org/10.1145/3355089.3356556, doi:10.1145/3355089.3356556

work page doi:10.1145/3355089.3356556 2019
[52]

Multimodal image synthesis and editing: The generative ai era

Zhan, F., Yu, Y., Wu, R., Zhang, J., Lu, S., Liu, L., Kortylewski, A., Theobalt, C., Xing, E., 2023. Multimodal image synthesis and editing: The generative ai era. URL:https://arxiv.org/abs/2112. 13592,arXiv:2112.13592

work page arXiv 2023
[53]

Adding conditional control to text-to-image diffusion models, in: Proceedings of the IEEE/CVF international conference on computer vision, pp

Zhang, L., Rao, A., Agrawala, M., 2023. Adding conditional control to text-to-image diffusion models, in: Proceedings of the IEEE/CVF international conference on computer vision, pp. 3836–3847

work page 2023
[54]

Scores as actions: a framework of fine-tuning diffusion models by continuous- time reinforcement learning

Zhao, H., Chen, H., Zhang, J., Yao, D.D., Tang, W., 2024. Scores as actions: a framework of fine-tuning diffusion models by continuous- time reinforcement learning. arXiv preprint arXiv:2409.08400

work page arXiv 2024
[55]

Struc- tured3d:Alargephoto-realisticdatasetforstructured3dmodeling,in: European Conference on Computer Vision, Springer

Zheng, J., Zhang, J., Li, J., Tang, R., Gao, S., Zhou, Z., 2020. Struc- tured3d:Alargephoto-realisticdatasetforstructured3dmodeling,in: European Conference on Computer Vision, Springer. pp. 519–535. :Preprint submitted to Elsevier Page 19 of 20

work page 2020
[56]

Neural-guided room layout generation with bubble diagram constraints

Zheng, Z., Petzold, F., 2023. Neural-guided room layout generation with bubble diagram constraints. Automation in construction 154, 104962. :Preprint submitted to Elsevier Page 20 of 20

work page 2023